Configuration

Three files, one precedence rule: config/local.yaml overrides config/settings.yaml, key by key. The Settings page writes local.yaml; never edit settings.yaml for personal values. config/skills.yaml is the active scoring profile (see Role presets).

Environment: JOBSCOUT_PORT runs a second instance (no scheduler); ADZUNA_APP_ID / ADZUNA_APP_KEY can replace the keys in local.yaml; MAX_THINKING_TOKENS is set by fit.py per call.

home

Key	Default	Meaning
city	Austin	your metro; postings matching city_terms → class home
city_terms	[austin, travis county]	lowercase substrings that mean the metro
state, state_code	Texas, TX	state name and code → class home_state
state_terms	DFW/Houston/SA metro cities	other metros in the state → home_state

filters

Key	Default	Meaning
max_age_days	45	drop aggregator postings older than this (when the age is known)
max_age_days_boards	60	same for direct boards; also hides older rows from the Inbox
keep_home_state	true	keep onsite jobs elsewhere in your state
keep_us_other_cities	false	keep onsite jobs in other US cities
keep_remote_global	false	keep remote jobs without a clear US signal

refresh

Key	Default	Meaning
at	"06:00, 11:00, 18:00, 23:00"	clock times (PC-local); "" = off
every_hours	0	interval fallback when at is empty
board_workers	8	parallel board fetches

adzuna

Key	Default	Meaning
enabled	true	skip Adzuna entirely when false
app_id, app_key	""	from developer.adzuna.com (free)
country	us
queries	12 role queries	each costs (locations + 1) calls per refresh
locations	home city + state metros	sweep locations
distance_km	60
companies, company_sweep_titles, company_sweep_days	12 category-3 employers	per-company title sweeps
max_days_old, results_per_page, pages_per_query	30, 50, 1	keep pages_per_query low on the free tier

fit

Key	Default	Meaning
enabled	true	score gated new jobs after every refresh
model	claude-haiku-4-5	the scorer
profile_model	claude-sonnet-5	rewrites fit-profile.md from feedback
thinking	full	full / small / off (~90 / 45 / 30 s per batch of 10)
batch	10	jobs per call
parallel	2	concurrent calls
gate_min_score	60	keyword score a job needs before the model looks at it
max_per_run	300	cap per start / refresh
auto_profile_after	5	pending feedback items that trigger a profile update (0 = manual)
timeout_min	8	per call
description_chars	1800	description excerpt sent per job

resume

Key	Default	Meaning
output_dir	resume/output	<Company>/<Role-slug>/<jobId>/
file_name	Tailored_Resume.docx	your resume file name; cover letters use the same stem
claude_cmd	claude	or a full path
model	claude-fable-5-1	both phases
timeout_min	10	per phase
batch_max	50	jobs per Tailor click
parallel	2	concurrent runs (1-3)

apply

Key	Default	Meaning
auto_plan	false	plan automatically when a resume becomes ready
mode	prefill	default per plan; unattended submits approved exact matches
unattended_enabled	true	master switch
allowed_ats	greenhouse, ashby, lever	eligible for unattended (Lever is captcha-locked to prefill anyway)
profile	resume/profile.yaml
max_per_run	5	forms per /apply-queue run
pause_seconds	20	between forms
window	06:00-06:30	prefill window used by a scheduled task

notify

Key	Default	Meaning
enabled	false
discord_webhook_url	""	local.yaml only
min_score	50
max_items	15	lines per message
loc_classes	home, home_state, remote_us

server

port: 8765. The server binds 127.0.0.1 and ::1.

config/skills.yaml

Sections: preset, relevance (title_terms, desc_rescue_groups, min_groups, search_terms), view, role_titles, title_blockers, title_penalties, seniority_bonus, skills (cap_points, groups), freshness_bonus, source_points, location_points, experience_bonus, flags, fit_profile_default. Details in Scoring and fit.

config/companies.yaml, watchlist.yaml

companies: [{name, ats, slug}] and companies: [{name, careers, platform, austin}] + boards: (manual search links). Both are appended by the Companies page.